select raw.timestamp, cd.`count`
from (
  select ts.ts as timestamp,
    max(d.timestamp) as datumTs
  from (
      select '2008-04-27 8:00:00' + interval 20 * num minute as ts,
        '2008-04-27 8:00:00' + interval 20 * (num - 1) minute as pts
      from sequence
      where num between 0 and 12
    ) ts
    left outer join (
      select timestamp
      from count_datum
      where seriesId = 47
    ) d on d.timestamp > ts.pts
      and d.timestamp <= ts.ts
  group by ts.ts
) raw
left outer join count_datum cd on cd.seriesId = 47
  and cd.timestamp = raw.datumTs

That's certainly faster that the double comparison, but noticeably slower than the single comparison, and it doesn't do quite the same thing. It returns the maximum `count` value for each window in time, not the last `count` value. E.g., for the '2009-04-27 11:00:00' record your query would return 8, not 6.

A different approach, though, so I'll play with it some more and see if I can get it to do the right thing.

Something like this? I didnt test this, so it might not even work.

select ts.ts as timestamp,       max(`counted`.`count`) as `count`
from (
    select '2009-04-27 8:00:00' + interval 20 * num minute as ts,
      '2009-04-27 8:00:00' + interval 20 * (num - 1) minute as pts
    from sequence
    where num between 0 and 12
  ) ts left join

  (
    select `count`, `timestamp`
    from count_datum
    where seriesId = 47
    order by timestamp desc
  ) as `counted` on
    `counted`.`timestamp` <= ts.ts and `counted`.`timestamp` > ts.pts
group by ts.ts